1 // SwiftShader Software Renderer
\r
3 // Copyright(c) 2005-2013 TransGaming Inc.
\r
5 // All rights reserved. No part of this software may be copied, distributed, transmitted,
\r
6 // transcribed, stored in a retrieval system, translated into any human or computer
\r
7 // language by any means, or disclosed to third parties without the explicit written
\r
8 // agreement of TransGaming Inc. Without such an agreement, no rights or licenses, express
\r
9 // or implied, including but not limited to any patent rights, are granted to you.
\r
12 #include "OutputASM.h"
\r
14 #include "common/debug.h"
\r
15 #include "InfoSink.h"
\r
17 #include "libGLESv2/Shader.h"
\r
20 #include <GLES2/gl2.h>
\r
21 #include <GLES2/gl2ext.h>
\r
22 #include <GLES3/gl3.h>
\r
26 // Integer to TString conversion
\r
30 sprintf(buffer, "%d", i);
\r
34 class Temporary : public TIntermSymbol
\r
37 Temporary(OutputASM *assembler) : TIntermSymbol(0, "tmp", TType(EbtFloat, EbpHigh, EvqTemporary, 4, 1, false)), assembler(assembler)
\r
43 assembler->freeTemporary(this);
\r
47 OutputASM *const assembler;
\r
50 class Constant : public TIntermConstantUnion
\r
53 Constant(float x, float y, float z, float w) : TIntermConstantUnion(constants, TType(EbtFloat, EbpHigh, EvqConstExpr, 4, 1, false))
\r
55 constants[0].setFConst(x);
\r
56 constants[1].setFConst(y);
\r
57 constants[2].setFConst(z);
\r
58 constants[3].setFConst(w);
\r
61 Constant(bool b) : TIntermConstantUnion(constants, TType(EbtBool, EbpHigh, EvqConstExpr, 1, 1, false))
\r
63 constants[0].setBConst(b);
\r
66 Constant(int i) : TIntermConstantUnion(constants, TType(EbtInt, EbpHigh, EvqConstExpr, 1, 1, false))
\r
68 constants[0].setIConst(i);
\r
76 ConstantUnion constants[4];
\r
79 Uniform::Uniform(GLenum type, GLenum precision, const std::string &name, int arraySize, int registerIndex)
\r
82 this->precision = precision;
\r
84 this->arraySize = arraySize;
\r
85 this->registerIndex = registerIndex;
\r
88 Attribute::Attribute()
\r
95 Attribute::Attribute(GLenum type, const std::string &name, int arraySize, int registerIndex)
\r
99 this->arraySize = arraySize;
\r
100 this->registerIndex = registerIndex;
\r
103 sw::PixelShader *Shader::getPixelShader() const
\r
108 sw::VertexShader *Shader::getVertexShader() const
\r
113 OutputASM::OutputASM(TParseContext &context, Shader *shaderObject) : TIntermTraverser(true, true, true), mContext(context), shaderObject(shaderObject)
\r
121 shader = shaderObject->getShader();
\r
122 pixelShader = shaderObject->getPixelShader();
\r
123 vertexShader = shaderObject->getVertexShader();
\r
126 functionArray.push_back(Function(0, "main(", 0, 0));
\r
127 currentFunction = 0;
\r
128 outputQualifier = EvqOutput; // Set outputQualifier to any value other than EvqFragColor or EvqFragData
\r
131 OutputASM::~OutputASM()
\r
135 void OutputASM::output()
\r
139 emitShader(GLOBAL);
\r
141 if(functionArray.size() > 1) // Only call main() when there are other functions
\r
143 Instruction *callMain = emit(sw::Shader::OPCODE_CALL);
\r
144 callMain->dst.type = sw::Shader::PARAMETER_LABEL;
\r
145 callMain->dst.index = 0; // main()
\r
147 emit(sw::Shader::OPCODE_RET);
\r
150 emitShader(FUNCTION);
\r
154 void OutputASM::emitShader(Scope scope)
\r
157 currentScope = GLOBAL;
\r
158 mContext.treeRoot->traverse(this);
\r
161 void OutputASM::freeTemporary(Temporary *temporary)
\r
163 free(temporaries, temporary);
\r
166 void OutputASM::visitSymbol(TIntermSymbol *symbol)
\r
168 // Vertex varyings don't have to be actively used to successfully link
\r
169 // against pixel shaders that use them. So make sure they're declared.
\r
170 if(symbol->getQualifier() == EvqVaryingOut || symbol->getQualifier() == EvqInvariantVaryingOut)
\r
172 if(symbol->getBasicType() != EbtInvariant) // Typeless declarations are not new varyings
\r
174 declareVarying(symbol, -1);
\r
179 bool OutputASM::visitBinary(Visit visit, TIntermBinary *node)
\r
181 if(currentScope != emitScope)
\r
186 TIntermTyped *result = node;
\r
187 TIntermTyped *left = node->getLeft();
\r
188 TIntermTyped *right = node->getRight();
\r
189 const TType &leftType = left->getType();
\r
190 const TType &rightType = right->getType();
\r
191 const TType &resultType = node->getType();
\r
193 switch(node->getOp())
\r
196 if(visit == PostVisit)
\r
198 assignLvalue(left, right);
\r
199 copy(result, right);
\r
202 case EOpInitialize:
\r
203 if(visit == PostVisit)
\r
208 case EOpMatrixTimesScalarAssign:
\r
209 if(visit == PostVisit)
\r
211 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
213 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
214 mul->dst.index += i;
\r
215 argument(mul->src[0], left, i);
\r
218 assignLvalue(left, result);
\r
221 case EOpVectorTimesMatrixAssign:
\r
222 if(visit == PostVisit)
\r
224 int size = leftType.getNominalSize();
\r
226 for(int i = 0; i < size; i++)
\r
228 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
229 dot->dst.mask = 1 << i;
\r
230 argument(dot->src[1], right, i);
\r
233 assignLvalue(left, result);
\r
236 case EOpMatrixTimesMatrixAssign:
\r
237 if(visit == PostVisit)
\r
239 int dim = leftType.getNominalSize();
\r
241 for(int i = 0; i < dim; i++)
\r
243 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
244 mul->dst.index += i;
\r
245 argument(mul->src[1], right, i);
\r
246 mul->src[1].swizzle = 0x00;
\r
248 for(int j = 1; j < dim; j++)
\r
250 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
251 mad->dst.index += i;
\r
252 argument(mad->src[0], left, j);
\r
253 argument(mad->src[1], right, i);
\r
254 mad->src[1].swizzle = j * 0x55;
\r
255 argument(mad->src[2], result, i);
\r
259 assignLvalue(left, result);
\r
262 case EOpIndexDirect:
\r
263 if(visit == PostVisit)
\r
265 int index = right->getAsConstantUnion()->getIConst(0);
\r
267 if(result->isMatrix() || result->isStruct())
\r
269 ASSERT(left->isArray());
\r
270 copy(result, left, index * left->elementRegisterCount());
\r
272 else if(result->isRegister())
\r
274 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
276 if(left->isRegister())
\r
278 mov->src[0].swizzle = index;
\r
280 else if(left->isArray())
\r
282 argument(mov->src[0], left, index * left->elementRegisterCount());
\r
284 else if(left->isMatrix())
\r
286 ASSERT(index < left->getNominalSize()); // FIXME: Report semantic error
\r
287 argument(mov->src[0], left, index);
\r
289 else UNREACHABLE();
\r
291 else UNREACHABLE();
\r
294 case EOpIndexIndirect:
\r
295 if(visit == PostVisit)
\r
297 if(left->isArray() || left->isMatrix())
\r
299 for(int index = 0; index < result->totalRegisterCount(); index++)
\r
301 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
302 mov->dst.index += index;
\r
303 mov->dst.mask = writeMask(result, index);
\r
304 argument(mov->src[0], left, index);
\r
306 if(left->totalRegisterCount() > 1)
\r
308 sw::Shader::SourceParameter relativeRegister;
\r
309 argument(relativeRegister, right);
\r
311 mov->src[0].rel.type = relativeRegister.type;
\r
312 mov->src[0].rel.index = relativeRegister.index;
\r
313 mov->src[0].rel.scale = result->totalRegisterCount();
\r
314 mov->src[0].rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
318 else if(left->isRegister())
\r
320 emit(sw::Shader::OPCODE_EXTRACT, result, left, right);
\r
322 else UNREACHABLE();
\r
325 case EOpIndexDirectStruct:
\r
326 if(visit == PostVisit)
\r
328 ASSERT(leftType.isStruct());
\r
330 const TTypeList *structure = leftType.getStruct();
\r
331 const TString &fieldName = rightType.getFieldName();
\r
332 int fieldOffset = 0;
\r
334 for(size_t i = 0; i < structure->size(); i++)
\r
336 const TType &fieldType = *(*structure)[i].type;
\r
338 if(fieldType.getFieldName() == fieldName)
\r
343 fieldOffset += fieldType.totalRegisterCount();
\r
346 copy(result, left, fieldOffset);
\r
349 case EOpVectorSwizzle:
\r
350 if(visit == PostVisit)
\r
353 TIntermAggregate *components = right->getAsAggregate();
\r
357 TIntermSequence &sequence = components->getSequence();
\r
360 for(TIntermSequence::iterator sit = sequence.begin(); sit != sequence.end(); sit++)
\r
362 TIntermConstantUnion *element = (*sit)->getAsConstantUnion();
\r
366 int i = element->getUnionArrayPointer()[0].getIConst();
\r
367 swizzle |= i << (component * 2);
\r
370 else UNREACHABLE();
\r
373 else UNREACHABLE();
\r
375 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, left);
\r
376 mov->src[0].swizzle = swizzle;
\r
379 case EOpAddAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_ADD, result, left, left, right); break;
\r
380 case EOpAdd: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_ADD, result, left, right); break;
\r
381 case EOpSubAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_SUB, result, left, left, right); break;
\r
382 case EOpSub: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_SUB, result, left, right); break;
\r
383 case EOpMulAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
384 case EOpMul: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
385 case EOpDivAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_DIV, result, left, left, right); break;
\r
386 case EOpDiv: if(visit == PostVisit) emitBinary(sw::Shader::OPCODE_DIV, result, left, right); break;
\r
388 if(visit == PostVisit)
\r
390 emitCmp(sw::Shader::CONTROL_EQ, result, left, right);
\r
392 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
394 Temporary equal(this);
\r
395 emitCmp(sw::Shader::CONTROL_EQ, &equal, left, right, index);
\r
396 emit(sw::Shader::OPCODE_AND, result, result, &equal);
\r
401 if(visit == PostVisit)
\r
403 emitCmp(sw::Shader::CONTROL_NE, result, left, right);
\r
405 for(int index = 1; index < left->totalRegisterCount(); index++)
\r
407 Temporary notEqual(this);
\r
408 emitCmp(sw::Shader::CONTROL_NE, ¬Equal, left, right, index);
\r
409 emit(sw::Shader::OPCODE_OR, result, result, ¬Equal);
\r
413 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, left, right); break;
\r
414 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, left, right); break;
\r
415 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, left, right); break;
\r
416 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, left, right); break;
\r
417 case EOpVectorTimesScalarAssign: if(visit == PostVisit) emitAssign(sw::Shader::OPCODE_MUL, result, left, left, right); break;
\r
418 case EOpVectorTimesScalar: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, left, right); break;
\r
419 case EOpMatrixTimesScalar:
\r
420 if(visit == PostVisit)
\r
422 for(int i = 0; i < leftType.getNominalSize(); i++)
\r
424 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
425 mul->dst.index += i;
\r
426 argument(mul->src[0], left, i);
\r
430 case EOpVectorTimesMatrix:
\r
431 if(visit == PostVisit)
\r
433 int size = leftType.getNominalSize();
\r
435 for(int i = 0; i < size; i++)
\r
437 Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
\r
438 dot->dst.mask = 1 << i;
\r
439 argument(dot->src[1], right, i);
\r
443 case EOpMatrixTimesVector:
\r
444 if(visit == PostVisit)
\r
446 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
447 mul->src[1].swizzle = 0x00;
\r
449 for(int i = 1; i < leftType.getNominalSize(); i++)
\r
451 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
452 argument(mad->src[0], left, i);
\r
453 mad->src[1].swizzle = i * 0x55;
\r
457 case EOpMatrixTimesMatrix:
\r
458 if(visit == PostVisit)
\r
460 int dim = leftType.getNominalSize();
\r
462 for(int i = 0; i < dim; i++)
\r
464 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
\r
465 mul->dst.index += i;
\r
466 argument(mul->src[1], right, i);
\r
467 mul->src[1].swizzle = 0x00;
\r
469 for(int j = 1; j < dim; j++)
\r
471 Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
\r
472 mad->dst.index += i;
\r
473 argument(mad->src[0], left, j);
\r
474 argument(mad->src[1], right, i);
\r
475 mad->src[1].swizzle = j * 0x55;
\r
476 argument(mad->src[2], result, i);
\r
482 if(trivial(right, 6))
\r
484 if(visit == PostVisit)
\r
486 emit(sw::Shader::OPCODE_OR, result, left, right);
\r
489 else // Short-circuit evaluation
\r
491 if(visit == InVisit)
\r
493 emit(sw::Shader::OPCODE_MOV, result, left);
\r
494 Instruction *ifnot = emit(sw::Shader::OPCODE_IF, 0, result);
\r
495 ifnot->src[0].modifier = sw::Shader::MODIFIER_NOT;
\r
497 else if(visit == PostVisit)
\r
499 emit(sw::Shader::OPCODE_MOV, result, right);
\r
500 emit(sw::Shader::OPCODE_ENDIF);
\r
504 case EOpLogicalXor: if(visit == PostVisit) emit(sw::Shader::OPCODE_XOR, result, left, right); break;
\r
505 case EOpLogicalAnd:
\r
506 if(trivial(right, 6))
\r
508 if(visit == PostVisit)
\r
510 emit(sw::Shader::OPCODE_AND, result, left, right);
\r
513 else // Short-circuit evaluation
\r
515 if(visit == InVisit)
\r
517 emit(sw::Shader::OPCODE_MOV, result, left);
\r
518 emit(sw::Shader::OPCODE_IF, 0, result);
\r
520 else if(visit == PostVisit)
\r
522 emit(sw::Shader::OPCODE_MOV, result, right);
\r
523 emit(sw::Shader::OPCODE_ENDIF);
\r
527 default: UNREACHABLE();
\r
533 bool OutputASM::visitUnary(Visit visit, TIntermUnary *node)
\r
535 if(currentScope != emitScope)
\r
540 Constant one(1.0f, 1.0f, 1.0f, 1.0f);
\r
541 Constant rad(1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f, 1.74532925e-2f);
\r
542 Constant deg(5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f, 5.72957795e+1f);
\r
544 TIntermTyped *result = node;
\r
545 TIntermTyped *arg = node->getOperand();
\r
547 switch(node->getOp())
\r
550 if(visit == PostVisit)
\r
552 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
554 Instruction *neg = emit(sw::Shader::OPCODE_MOV, result, arg);
\r
555 neg->dst.index += index;
\r
556 argument(neg->src[0], arg, index);
\r
557 neg->src[0].modifier = sw::Shader::MODIFIER_NEGATE;
\r
561 case EOpVectorLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
562 case EOpLogicalNot: if(visit == PostVisit) emit(sw::Shader::OPCODE_NOT, result, arg); break;
\r
563 case EOpPostIncrement:
\r
564 if(visit == PostVisit)
\r
568 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
570 Instruction *add = emit(sw::Shader::OPCODE_ADD, arg, arg, &one);
\r
571 add->dst.index += index;
\r
572 argument(add->src[0], arg, index);
\r
575 assignLvalue(arg, arg);
\r
578 case EOpPostDecrement:
\r
579 if(visit == PostVisit)
\r
583 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
585 Instruction *sub = emit(sw::Shader::OPCODE_SUB, arg, arg, &one);
\r
586 sub->dst.index += index;
\r
587 argument(sub->src[0], arg, index);
\r
590 assignLvalue(arg, arg);
\r
593 case EOpPreIncrement:
\r
594 if(visit == PostVisit)
\r
596 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
598 Instruction *add = emit(sw::Shader::OPCODE_ADD, result, arg, &one);
\r
599 add->dst.index += index;
\r
600 argument(add->src[0], arg, index);
\r
603 assignLvalue(arg, result);
\r
606 case EOpPreDecrement:
\r
607 if(visit == PostVisit)
\r
609 for(int index = 0; index < arg->totalRegisterCount(); index++)
\r
611 Instruction *sub = emit(sw::Shader::OPCODE_SUB, result, arg, &one);
\r
612 sub->dst.index += index;
\r
613 argument(sub->src[0], arg, index);
\r
616 assignLvalue(arg, result);
\r
619 case EOpRadians: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, &rad); break;
\r
620 case EOpDegrees: if(visit == PostVisit) emit(sw::Shader::OPCODE_MUL, result, arg, °); break;
\r
621 case EOpSin: if(visit == PostVisit) emit(sw::Shader::OPCODE_SIN, result, arg); break;
\r
622 case EOpCos: if(visit == PostVisit) emit(sw::Shader::OPCODE_COS, result, arg); break;
\r
623 case EOpTan: if(visit == PostVisit) emit(sw::Shader::OPCODE_TAN, result, arg); break;
\r
624 case EOpAsin: if(visit == PostVisit) emit(sw::Shader::OPCODE_ASIN, result, arg); break;
\r
625 case EOpAcos: if(visit == PostVisit) emit(sw::Shader::OPCODE_ACOS, result, arg); break;
\r
626 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN, result, arg); break;
\r
627 case EOpExp: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP, result, arg); break;
\r
628 case EOpLog: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG, result, arg); break;
\r
629 case EOpExp2: if(visit == PostVisit) emit(sw::Shader::OPCODE_EXP2, result, arg); break;
\r
630 case EOpLog2: if(visit == PostVisit) emit(sw::Shader::OPCODE_LOG2, result, arg); break;
\r
631 case EOpSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_SQRT, result, arg); break;
\r
632 case EOpInverseSqrt: if(visit == PostVisit) emit(sw::Shader::OPCODE_RSQ, result, arg); break;
\r
633 case EOpAbs: if(visit == PostVisit) emit(sw::Shader::OPCODE_ABS, result, arg); break;
\r
634 case EOpSign: if(visit == PostVisit) emit(sw::Shader::OPCODE_SGN, result, arg); break;
\r
635 case EOpFloor: if(visit == PostVisit) emit(sw::Shader::OPCODE_FLOOR, result, arg); break;
\r
636 case EOpCeil: if(visit == PostVisit) emit(sw::Shader::OPCODE_CEIL, result, arg, result); break;
\r
637 case EOpFract: if(visit == PostVisit) emit(sw::Shader::OPCODE_FRC, result, arg); break;
\r
638 case EOpLength: if(visit == PostVisit) emit(sw::Shader::OPCODE_LEN(dim(arg)), result, arg); break;
\r
639 case EOpNormalize: if(visit == PostVisit) emit(sw::Shader::OPCODE_NRM(dim(arg)), result, arg); break;
\r
640 case EOpDFdx: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDX, result, arg); break;
\r
641 case EOpDFdy: if(visit == PostVisit) emit(sw::Shader::OPCODE_DFDY, result, arg); break;
\r
642 case EOpFwidth: if(visit == PostVisit) emit(sw::Shader::OPCODE_FWIDTH, result, arg); break;
\r
643 case EOpAny: if(visit == PostVisit) emit(sw::Shader::OPCODE_ANY, result, arg); break;
\r
644 case EOpAll: if(visit == PostVisit) emit(sw::Shader::OPCODE_ALL, result, arg); break;
\r
645 default: UNREACHABLE();
\r
651 bool OutputASM::visitAggregate(Visit visit, TIntermAggregate *node)
\r
653 if(currentScope != emitScope && node->getOp() != EOpFunction && node->getOp() != EOpSequence)
\r
658 Constant zero(0.0f, 0.0f, 0.0f, 0.0f);
\r
660 TIntermTyped *result = node;
\r
661 const TType &resultType = node->getType();
\r
662 TIntermSequence &arg = node->getSequence();
\r
663 int argumentCount = arg.size();
\r
665 switch(node->getOp())
\r
667 case EOpSequence: break;
\r
668 case EOpDeclaration: break;
\r
669 case EOpPrototype: break;
\r
671 if(visit == PostVisit)
\r
673 copy(result, arg[1]);
\r
677 if(visit == PreVisit)
\r
679 const TString &name = node->getName();
\r
681 if(emitScope == FUNCTION)
\r
683 if(functionArray.size() > 1) // No need for a label when there's only main()
\r
685 Instruction *label = emit(sw::Shader::OPCODE_LABEL);
\r
686 label->dst.type = sw::Shader::PARAMETER_LABEL;
\r
688 const Function *function = findFunction(name);
\r
689 ASSERT(function); // Should have been added during global pass
\r
690 label->dst.index = function->label;
\r
691 currentFunction = function->label;
\r
694 else if(emitScope == GLOBAL)
\r
696 if(name != "main(")
\r
698 TIntermSequence &arguments = node->getSequence()[0]->getAsAggregate()->getSequence();
\r
699 functionArray.push_back(Function(functionArray.size(), name, &arguments, node));
\r
702 else UNREACHABLE();
\r
704 currentScope = FUNCTION;
\r
706 else if(visit == PostVisit)
\r
708 if(emitScope == FUNCTION)
\r
710 if(functionArray.size() > 1) // No need to return when there's only main()
\r
712 emit(sw::Shader::OPCODE_RET);
\r
716 currentScope = GLOBAL;
\r
719 case EOpFunctionCall:
\r
720 if(visit == PostVisit)
\r
722 if(node->isUserDefined())
\r
724 const TString &name = node->getName();
\r
725 const Function *function = findFunction(name);
\r
729 mContext.error(node->getLine(), "function definition not found", name.c_str());
\r
733 TIntermSequence &arguments = *function->arg;
\r
735 for(int i = 0; i < argumentCount; i++)
\r
737 TIntermTyped *in = arguments[i]->getAsTyped();
\r
739 if(in->getQualifier() == EvqIn ||
\r
740 in->getQualifier() == EvqInOut ||
\r
741 in->getQualifier() == EvqConstReadOnly)
\r
747 Instruction *call = emit(sw::Shader::OPCODE_CALL);
\r
748 call->dst.type = sw::Shader::PARAMETER_LABEL;
\r
749 call->dst.index = function->label;
\r
751 if(function->ret && function->ret->getType().getBasicType() != EbtVoid)
\r
753 copy(result, function->ret);
\r
756 for(int i = 0; i < argumentCount; i++)
\r
758 TIntermTyped *argument = arguments[i]->getAsTyped();
\r
759 TIntermTyped *out = arg[i]->getAsTyped();
\r
761 if(argument->getQualifier() == EvqOut ||
\r
762 argument->getQualifier() == EvqInOut)
\r
764 copy(out, argument);
\r
770 TString name = TFunction::unmangleName(node->getName());
\r
772 if(name == "texture" || name == "texture2D" || name == "textureCube" || name == "texture3D")
\r
774 if(argumentCount == 2)
\r
776 emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
778 else if(argumentCount == 3) // bias
\r
780 Temporary uvwb(this);
\r
781 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
782 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
783 bias->dst.mask = 0x8;
\r
785 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &uvwb, arg[0]); // FIXME: Implement an efficient TEXLDB instruction
\r
788 else UNREACHABLE();
\r
790 else if(name == "texture2DProj")
\r
792 TIntermTyped *t = arg[1]->getAsTyped();
\r
794 if(argumentCount == 2)
\r
796 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, arg[1], arg[0]);
\r
797 tex->project = true;
\r
799 if(t->getNominalSize() == 3)
\r
801 tex->src[0].swizzle = 0xA4;
\r
803 else ASSERT(t->getNominalSize() == 4);
\r
805 else if(argumentCount == 3) // bias
\r
807 Temporary proj(this);
\r
809 if(t->getNominalSize() == 3)
\r
811 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
812 div->src[1].swizzle = 0xAA;
\r
813 div->dst.mask = 0x3;
\r
815 else if(t->getNominalSize() == 4)
\r
817 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
818 div->src[1].swizzle = 0xFF;
\r
819 div->dst.mask = 0x3;
\r
821 else UNREACHABLE();
\r
823 Instruction *bias = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
824 bias->dst.mask = 0x8;
\r
826 Instruction *tex = emit(sw::Shader::OPCODE_TEX, result, &proj, arg[0]);
\r
829 else UNREACHABLE();
\r
831 else if(name == "texture2DLod" || name == "textureCubeLod")
\r
833 Temporary uvwb(this);
\r
834 emit(sw::Shader::OPCODE_MOV, &uvwb, arg[1]);
\r
835 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &uvwb, arg[2]);
\r
836 lod->dst.mask = 0x8;
\r
838 emit(sw::Shader::OPCODE_TEXLDL, result, &uvwb, arg[0]);
\r
840 else if(name == "texture2DProjLod")
\r
842 TIntermTyped *t = arg[1]->getAsTyped();
\r
843 Temporary proj(this);
\r
845 if(t->getNominalSize() == 3)
\r
847 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
848 div->src[1].swizzle = 0xAA;
\r
849 div->dst.mask = 0x3;
\r
851 else if(t->getNominalSize() == 4)
\r
853 Instruction *div = emit(sw::Shader::OPCODE_DIV, &proj, arg[1], arg[1]);
\r
854 div->src[1].swizzle = 0xFF;
\r
855 div->dst.mask = 0x3;
\r
857 else UNREACHABLE();
\r
859 Instruction *lod = emit(sw::Shader::OPCODE_MOV, &proj, arg[2]);
\r
860 lod->dst.mask = 0x8;
\r
862 emit(sw::Shader::OPCODE_TEXLDL, result, &proj, arg[0]);
\r
864 else UNREACHABLE();
\r
868 case EOpParameters:
\r
870 case EOpConstructFloat:
\r
871 case EOpConstructVec2:
\r
872 case EOpConstructVec3:
\r
873 case EOpConstructVec4:
\r
874 case EOpConstructBool:
\r
875 case EOpConstructBVec2:
\r
876 case EOpConstructBVec3:
\r
877 case EOpConstructBVec4:
\r
878 case EOpConstructInt:
\r
879 case EOpConstructIVec2:
\r
880 case EOpConstructIVec3:
\r
881 case EOpConstructIVec4:
\r
882 if(visit == PostVisit)
\r
886 for(int i = 0; i < argumentCount; i++)
\r
888 TIntermTyped *argi = arg[i]->getAsTyped();
\r
889 int size = argi->getNominalSize();
\r
891 if(!argi->isMatrix())
\r
893 Instruction *mov = emitCast(result, argi);
\r
894 mov->dst.mask = (0xF << component) & 0xF;
\r
895 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
903 while(component < resultType.getNominalSize())
\r
905 Instruction *mov = emitCast(result, argi);
\r
906 mov->dst.mask = (0xF << component) & 0xF;
\r
907 mov->src[0].index += column;
\r
908 mov->src[0].swizzle = readSwizzle(argi, size) << (component * 2);
\r
917 case EOpConstructMat2:
\r
918 case EOpConstructMat3:
\r
919 case EOpConstructMat4:
\r
920 if(visit == PostVisit)
\r
922 TIntermTyped *arg0 = arg[0]->getAsTyped();
\r
923 const int dim = result->getNominalSize();
\r
925 if(arg0->isScalar() && arg.size() == 1) // Construct scale matrix
\r
927 for(int i = 0; i < dim; i++)
\r
929 Instruction *init = emit(sw::Shader::OPCODE_MOV, result, &zero);
\r
930 init->dst.index += i;
\r
931 Instruction *mov = emitCast(result, arg0);
\r
932 mov->dst.index += i;
\r
933 mov->dst.mask = 1 << i;
\r
934 ASSERT(mov->src[0].swizzle == 0x00);
\r
937 else if(arg0->isMatrix())
\r
939 for(int i = 0; i < dim; i++)
\r
941 if(dim > dim2(arg0))
\r
943 // Initialize to identity matrix
\r
944 Constant col((i == 0 ? 1.0f : 0.0f), (i == 1 ? 1.0f : 0.0f), (i == 2 ? 1.0f : 0.0f), (i == 3 ? 1.0f : 0.0f));
\r
945 Instruction *mov = emitCast(result, &col);
\r
946 mov->dst.index += i;
\r
951 Instruction *mov = emitCast(result, arg0);
\r
952 mov->dst.index += i;
\r
953 mov->dst.mask = 0xF >> (4 - dim2(arg0));
\r
954 argument(mov->src[0], arg0, i);
\r
963 for(int i = 0; i < argumentCount; i++)
\r
965 TIntermTyped *argi = arg[i]->getAsTyped();
\r
966 int size = argi->getNominalSize();
\r
969 while(element < size)
\r
971 Instruction *mov = emitCast(result, argi);
\r
972 mov->dst.index += column;
\r
973 mov->dst.mask = (0xF << row) & 0xF;
\r
974 mov->src[0].swizzle = (readSwizzle(argi, size) << (row * 2)) + 0x55 * element;
\r
976 int end = row + size - element;
\r
977 column = end >= dim ? column + 1 : column;
\r
978 element = element + dim - row;
\r
979 row = end >= dim ? 0 : end;
\r
985 case EOpConstructStruct:
\r
986 if(visit == PostVisit)
\r
989 for(int i = 0; i < argumentCount; i++)
\r
991 TIntermTyped *argi = arg[i]->getAsTyped();
\r
992 int size = argi->totalRegisterCount();
\r
994 for(int index = 0; index < size; index++)
\r
996 Instruction *mov = emit(sw::Shader::OPCODE_MOV, result, argi);
\r
997 mov->dst.index += index + offset;
\r
998 mov->dst.mask = writeMask(result, offset + index);
\r
999 argument(mov->src[0], argi, index);
\r
1006 case EOpLessThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LT, result, arg[0], arg[1]); break;
\r
1007 case EOpGreaterThan: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GT, result, arg[0], arg[1]); break;
\r
1008 case EOpLessThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_LE, result, arg[0], arg[1]); break;
\r
1009 case EOpGreaterThanEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_GE, result, arg[0], arg[1]); break;
\r
1010 case EOpVectorEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_EQ, result, arg[0], arg[1]); break;
\r
1011 case EOpVectorNotEqual: if(visit == PostVisit) emitCmp(sw::Shader::CONTROL_NE, result, arg[0], arg[1]); break;
\r
1012 case EOpMod: if(visit == PostVisit) emit(sw::Shader::OPCODE_MOD, result, arg[0], arg[1]); break;
\r
1013 case EOpPow: if(visit == PostVisit) emit(sw::Shader::OPCODE_POW, result, arg[0], arg[1]); break;
\r
1014 case EOpAtan: if(visit == PostVisit) emit(sw::Shader::OPCODE_ATAN2, result, arg[0], arg[1]); break;
\r
1015 case EOpMin: if(visit == PostVisit) emit(sw::Shader::OPCODE_MIN, result, arg[0], arg[1]); break;
\r
1016 case EOpMax: if(visit == PostVisit) emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]); break;
\r
1018 if(visit == PostVisit)
\r
1020 emit(sw::Shader::OPCODE_MAX, result, arg[0], arg[1]);
\r
1021 emit(sw::Shader::OPCODE_MIN, result, result, arg[2]);
\r
1024 case EOpMix: if(visit == PostVisit) emit(sw::Shader::OPCODE_LRP, result, arg[2], arg[1], arg[0]); break;
\r
1025 case EOpStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_STEP, result, arg[0], arg[1]); break;
\r
1026 case EOpSmoothStep: if(visit == PostVisit) emit(sw::Shader::OPCODE_SMOOTH, result, arg[0], arg[1], arg[2]); break;
\r
1027 case EOpDistance: if(visit == PostVisit) emit(sw::Shader::OPCODE_DIST(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1028 case EOpDot: if(visit == PostVisit) emit(sw::Shader::OPCODE_DP(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1029 case EOpCross: if(visit == PostVisit) emit(sw::Shader::OPCODE_CRS, result, arg[0], arg[1]); break;
\r
1030 case EOpFaceForward: if(visit == PostVisit) emit(sw::Shader::OPCODE_FORWARD(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1031 case EOpReflect: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFLECT(dim(arg[0])), result, arg[0], arg[1]); break;
\r
1032 case EOpRefract: if(visit == PostVisit) emit(sw::Shader::OPCODE_REFRACT(dim(arg[0])), result, arg[0], arg[1], arg[2]); break;
\r
1034 if(visit == PostVisit)
\r
1036 ASSERT(dim2(arg[0]) == dim2(arg[1]));
\r
1038 for(int i = 0; i < dim2(arg[0]); i++)
\r
1040 Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
\r
1041 mul->dst.index += i;
\r
1042 argument(mul->src[0], arg[0], i);
\r
1043 argument(mul->src[1], arg[1], i);
\r
1047 default: UNREACHABLE();
\r
1053 bool OutputASM::visitSelection(Visit visit, TIntermSelection *node)
\r
1055 if(currentScope != emitScope)
\r
1060 TIntermTyped *condition = node->getCondition();
\r
1061 TIntermNode *trueBlock = node->getTrueBlock();
\r
1062 TIntermNode *falseBlock = node->getFalseBlock();
\r
1063 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1065 condition->traverse(this);
\r
1067 if(node->usesTernaryOperator())
\r
1069 if(constantCondition)
\r
1071 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1075 trueBlock->traverse(this);
\r
1076 copy(node, trueBlock);
\r
1080 falseBlock->traverse(this);
\r
1081 copy(node, falseBlock);
\r
1084 else if(trivial(node, 6)) // Fast to compute both potential results and no side effects
\r
1086 trueBlock->traverse(this);
\r
1087 falseBlock->traverse(this);
\r
1088 emit(sw::Shader::OPCODE_SELECT, node, condition, trueBlock, falseBlock);
\r
1092 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1096 trueBlock->traverse(this);
\r
1097 copy(node, trueBlock);
\r
1102 emit(sw::Shader::OPCODE_ELSE);
\r
1103 falseBlock->traverse(this);
\r
1104 copy(node, falseBlock);
\r
1107 emit(sw::Shader::OPCODE_ENDIF);
\r
1110 else // if/else statement
\r
1112 if(constantCondition)
\r
1114 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1120 trueBlock->traverse(this);
\r
1127 falseBlock->traverse(this);
\r
1133 emit(sw::Shader::OPCODE_IF, 0, condition);
\r
1137 trueBlock->traverse(this);
\r
1142 emit(sw::Shader::OPCODE_ELSE);
\r
1143 falseBlock->traverse(this);
\r
1146 emit(sw::Shader::OPCODE_ENDIF);
\r
1153 bool OutputASM::visitLoop(Visit visit, TIntermLoop *node)
\r
1155 if(currentScope != emitScope)
\r
1160 unsigned int iterations = loopCount(node);
\r
1162 if(iterations == 0)
\r
1167 bool unroll = (iterations <= 4);
\r
1171 DetectLoopDiscontinuity detectLoopDiscontinuity;
\r
1172 unroll = !detectLoopDiscontinuity.traverse(node);
\r
1175 TIntermNode *init = node->getInit();
\r
1176 TIntermTyped *condition = node->getCondition();
\r
1177 TIntermTyped *expression = node->getExpression();
\r
1178 TIntermNode *body = node->getBody();
\r
1180 if(node->getType() == ELoopDoWhile)
\r
1182 Temporary iterate(this);
\r
1183 Constant True(true);
\r
1184 emit(sw::Shader::OPCODE_MOV, &iterate, &True);
\r
1186 emit(sw::Shader::OPCODE_WHILE, 0, &iterate); // FIXME: Implement real do-while
\r
1190 body->traverse(this);
\r
1193 emit(sw::Shader::OPCODE_TEST);
\r
1195 condition->traverse(this);
\r
1196 emit(sw::Shader::OPCODE_MOV, &iterate, condition);
\r
1198 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1204 init->traverse(this);
\r
1209 for(unsigned int i = 0; i < iterations; i++)
\r
1211 // condition->traverse(this); // Condition could contain statements, but not in an unrollable loop
\r
1215 body->traverse(this);
\r
1220 expression->traverse(this);
\r
1226 condition->traverse(this);
\r
1228 emit(sw::Shader::OPCODE_WHILE, 0, condition);
\r
1232 body->traverse(this);
\r
1235 emit(sw::Shader::OPCODE_TEST);
\r
1239 expression->traverse(this);
\r
1242 condition->traverse(this);
\r
1244 emit(sw::Shader::OPCODE_ENDWHILE);
\r
1251 bool OutputASM::visitBranch(Visit visit, TIntermBranch *node)
\r
1253 if(currentScope != emitScope)
\r
1258 switch(node->getFlowOp())
\r
1260 case EOpKill: if(visit == PostVisit) emit(sw::Shader::OPCODE_DISCARD); break;
\r
1261 case EOpBreak: if(visit == PostVisit) emit(sw::Shader::OPCODE_BREAK); break;
\r
1262 case EOpContinue: if(visit == PostVisit) emit(sw::Shader::OPCODE_CONTINUE); break;
\r
1264 if(visit == PostVisit)
\r
1266 TIntermTyped *value = node->getExpression();
\r
1270 copy(functionArray[currentFunction].ret, value);
\r
1273 emit(sw::Shader::OPCODE_LEAVE);
\r
1276 default: UNREACHABLE();
\r
1282 bool OutputASM::isSamplerRegister(TIntermTyped *operand)
\r
1284 return operand && isSamplerRegister(operand->getType());
\r
1287 bool OutputASM::isSamplerRegister(const TType &type)
\r
1289 // A sampler register's qualifiers can be:
\r
1290 // - EvqUniform: The sampler uniform is used as is in the code (default case).
\r
1291 // - EvqTemporary: The sampler is indexed. It's still a sampler register.
\r
1292 // - EvqIn (and other similar types): The sampler has been passed as a function argument. At this point,
\r
1293 // the sampler has been copied and is no longer a sampler register.
\r
1294 return IsSampler(type.getBasicType()) && (type.getQualifier() == EvqUniform || type.getQualifier() == EvqTemporary);
\r
1297 Instruction *OutputASM::emit(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2, int index)
\r
1299 if(isSamplerRegister(dst))
\r
1301 op = sw::Shader::OPCODE_NULL; // Can't assign to a sampler, but this is hit when indexing sampler arrays
\r
1304 Instruction *instruction = new Instruction(op);
\r
1308 instruction->dst.type = registerType(dst);
\r
1309 instruction->dst.index = registerIndex(dst) + index;
\r
1310 instruction->dst.mask = writeMask(dst);
\r
1311 instruction->dst.integer = (dst->getBasicType() == EbtInt);
\r
1314 argument(instruction->src[0], src0, index);
\r
1315 argument(instruction->src[1], src1, index);
\r
1316 argument(instruction->src[2], src2, index);
\r
1318 shader->append(instruction);
\r
1320 return instruction;
\r
1323 Instruction *OutputASM::emitCast(TIntermTyped *dst, TIntermTyped *src)
\r
1325 // Integers are implemented as float
\r
1326 if((dst->getBasicType() == EbtFloat || dst->getBasicType() == EbtInt) && src->getBasicType() == EbtBool)
\r
1328 return emit(sw::Shader::OPCODE_B2F, dst, src);
\r
1330 if(dst->getBasicType() == EbtBool && (src->getBasicType() == EbtFloat || src->getBasicType() == EbtInt))
\r
1332 return emit(sw::Shader::OPCODE_F2B, dst, src);
\r
1334 if(dst->getBasicType() == EbtInt && src->getBasicType() == EbtFloat)
\r
1336 return emit(sw::Shader::OPCODE_TRUNC, dst, src);
\r
1339 return emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1342 void OutputASM::emitBinary(sw::Shader::Opcode op, TIntermTyped *dst, TIntermNode *src0, TIntermNode *src1, TIntermNode *src2)
\r
1344 for(int index = 0; index < dst->elementRegisterCount(); index++)
\r
1346 emit(op, dst, src0, src1, src2, index);
\r
1350 void OutputASM::emitAssign(sw::Shader::Opcode op, TIntermTyped *result, TIntermTyped *lhs, TIntermTyped *src0, TIntermTyped *src1)
\r
1352 emitBinary(op, result, src0, src1);
\r
1353 assignLvalue(lhs, result);
\r
1356 void OutputASM::emitCmp(sw::Shader::Control cmpOp, TIntermTyped *dst, TIntermNode *left, TIntermNode *right, int index)
\r
1358 bool boolean = (left->getAsTyped()->getBasicType() == EbtBool);
\r
1359 sw::Shader::Opcode opcode = boolean ? sw::Shader::OPCODE_ICMP : sw::Shader::OPCODE_CMP;
\r
1361 Instruction *cmp = emit(opcode, dst, left, right);
\r
1362 cmp->control = cmpOp;
\r
1363 argument(cmp->src[0], left, index);
\r
1364 argument(cmp->src[1], right, index);
\r
1367 int componentCount(const TType &type, int registers)
\r
1369 if(registers == 0)
\r
1374 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1376 int index = registers / type.elementRegisterCount();
\r
1377 registers -= index * type.elementRegisterCount();
\r
1378 return index * type.getElementSize() + componentCount(type, registers);
\r
1381 if(type.isStruct())
\r
1383 TTypeList *structure = type.getStruct();
\r
1386 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1388 const TType &fieldType = *field->type;
\r
1390 if(fieldType.totalRegisterCount() <= registers)
\r
1392 registers -= fieldType.totalRegisterCount();
\r
1393 elements += fieldType.getObjectSize();
\r
1395 else // Register within this field
\r
1397 return elements + componentCount(fieldType, registers);
\r
1401 else if(type.isMatrix())
\r
1403 return registers * type.getNominalSize();
\r
1410 int registerSize(const TType &type, int registers)
\r
1412 if(registers == 0)
\r
1414 if(type.isStruct())
\r
1416 return registerSize(*type.getStruct()->begin()->type, 0);
\r
1419 return type.getNominalSize();
\r
1422 if(type.isArray() && registers >= type.elementRegisterCount())
\r
1424 int index = registers / type.elementRegisterCount();
\r
1425 registers -= index * type.elementRegisterCount();
\r
1426 return registerSize(type, registers);
\r
1429 if(type.isStruct())
\r
1431 TTypeList *structure = type.getStruct();
\r
1434 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1436 const TType &fieldType = *field->type;
\r
1438 if(fieldType.totalRegisterCount() <= registers)
\r
1440 registers -= fieldType.totalRegisterCount();
\r
1441 elements += fieldType.getObjectSize();
\r
1443 else // Register within this field
\r
1445 return registerSize(fieldType, registers);
\r
1449 else if(type.isMatrix())
\r
1451 return registerSize(type, 0);
\r
1458 void OutputASM::argument(sw::Shader::SourceParameter ¶meter, TIntermNode *argument, int index)
\r
1462 TIntermTyped *arg = argument->getAsTyped();
\r
1463 const TType &type = arg->getType();
\r
1464 const TTypeList *structure = type.getStruct();
\r
1465 index = (index >= arg->totalRegisterCount()) ? arg->totalRegisterCount() - 1 : index;
\r
1467 int size = registerSize(type, index);
\r
1469 parameter.type = registerType(arg);
\r
1471 if(arg->getQualifier() == EvqConstExpr)
\r
1473 int component = componentCount(type, index);
\r
1474 ConstantUnion *constants = arg->getAsConstantUnion()->getUnionArrayPointer();
\r
1476 for(int i = 0; i < 4; i++)
\r
1478 if(size == 1) // Replicate
\r
1480 parameter.value[i] = constants[component + 0].getAsFloat();
\r
1484 parameter.value[i] = constants[component + i].getAsFloat();
\r
1488 parameter.value[i] = 0.0f;
\r
1494 parameter.index = registerIndex(arg) + index;
\r
1496 if(isSamplerRegister(arg))
\r
1498 TIntermBinary *binary = argument->getAsBinaryNode();
\r
1502 TIntermTyped *left = binary->getLeft();
\r
1503 TIntermTyped *right = binary->getRight();
\r
1505 if(binary->getOp() == EOpIndexDirect)
\r
1507 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1509 else if(binary->getOp() == EOpIndexIndirect)
\r
1511 if(left->getArraySize() > 1)
\r
1513 parameter.rel.type = registerType(binary->getRight());
\r
1514 parameter.rel.index = registerIndex(binary->getRight());
\r
1515 parameter.rel.scale = 1;
\r
1516 parameter.rel.deterministic = true;
\r
1519 else if(binary->getOp() == EOpIndexDirectStruct)
\r
1521 parameter.index += right->getAsConstantUnion()->getIConst(0);
\r
1523 else UNREACHABLE();
\r
1528 if(!IsSampler(arg->getBasicType()))
\r
1530 parameter.swizzle = readSwizzle(arg, size);
\r
1535 void OutputASM::copy(TIntermTyped *dst, TIntermNode *src, int offset)
\r
1537 for(int index = 0; index < dst->totalRegisterCount(); index++)
\r
1539 Instruction *mov = emit(sw::Shader::OPCODE_MOV, dst, src);
\r
1540 mov->dst.index += index;
\r
1541 mov->dst.mask = writeMask(dst, index);
\r
1542 argument(mov->src[0], src, offset + index);
\r
1546 int swizzleElement(int swizzle, int index)
\r
1548 return (swizzle >> (index * 2)) & 0x03;
\r
1551 int swizzleSwizzle(int leftSwizzle, int rightSwizzle)
\r
1553 return (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 0)) << 0) |
\r
1554 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 1)) << 2) |
\r
1555 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 2)) << 4) |
\r
1556 (swizzleElement(leftSwizzle, swizzleElement(rightSwizzle, 3)) << 6);
\r
1559 void OutputASM::assignLvalue(TIntermTyped *dst, TIntermTyped *src)
\r
1562 ((src->isVector() && (!dst->isVector() || (dst->getNominalSize() != dst->getNominalSize()))) ||
\r
1563 (src->isMatrix() && (!dst->isMatrix() || (src->getNominalSize() != dst->getNominalSize())))))
\r
1565 return mContext.error(src->getLine(), "Result type should match the l-value type in compound assignment", src->isVector() ? "vector" : "matrix");
\r
1568 TIntermBinary *binary = dst->getAsBinaryNode();
\r
1570 if(binary && binary->getOp() == EOpIndexIndirect && dst->isScalar())
\r
1572 Instruction *insert = new Instruction(sw::Shader::OPCODE_INSERT);
\r
1574 Temporary address(this);
\r
1575 lvalue(insert->dst, address, dst);
\r
1577 insert->src[0].type = insert->dst.type;
\r
1578 insert->src[0].index = insert->dst.index;
\r
1579 insert->src[0].rel = insert->dst.rel;
\r
1580 argument(insert->src[1], src);
\r
1581 argument(insert->src[2], binary->getRight());
\r
1583 shader->append(insert);
\r
1587 for(int offset = 0; offset < dst->totalRegisterCount(); offset++)
\r
1589 Instruction *mov = new Instruction(sw::Shader::OPCODE_MOV);
\r
1591 Temporary address(this);
\r
1592 int swizzle = lvalue(mov->dst, address, dst);
\r
1593 mov->dst.index += offset;
\r
1597 mov->dst.mask = writeMask(dst, offset);
\r
1600 argument(mov->src[0], src, offset);
\r
1601 mov->src[0].swizzle = swizzleSwizzle(mov->src[0].swizzle, swizzle);
\r
1603 shader->append(mov);
\r
1608 int OutputASM::lvalue(sw::Shader::DestinationParameter &dst, Temporary &address, TIntermTyped *node)
\r
1610 TIntermTyped *result = node;
\r
1611 TIntermBinary *binary = node->getAsBinaryNode();
\r
1612 TIntermSymbol *symbol = node->getAsSymbolNode();
\r
1616 TIntermTyped *left = binary->getLeft();
\r
1617 TIntermTyped *right = binary->getRight();
\r
1619 int leftSwizzle = lvalue(dst, address, left); // Resolve the l-value of the left side
\r
1621 switch(binary->getOp())
\r
1623 case EOpIndexDirect:
\r
1625 int rightIndex = right->getAsConstantUnion()->getIConst(0);
\r
1627 if(left->isRegister())
\r
1629 int leftMask = dst.mask;
\r
1632 while((leftMask & dst.mask) == 0)
\r
1634 dst.mask = dst.mask << 1;
\r
1637 int element = swizzleElement(leftSwizzle, rightIndex);
\r
1638 dst.mask = 1 << element;
\r
1642 else if(left->isArray() || left->isMatrix())
\r
1644 dst.index += rightIndex * result->totalRegisterCount();
\r
1647 else UNREACHABLE();
\r
1650 case EOpIndexIndirect:
\r
1652 if(left->isRegister())
\r
1654 // Requires INSERT instruction (handled by calling function)
\r
1656 else if(left->isArray() || left->isMatrix())
\r
1658 int scale = result->totalRegisterCount();
\r
1660 if(dst.rel.type == sw::Shader::PARAMETER_VOID) // Use the index register as the relative address directly
\r
1662 if(left->totalRegisterCount() > 1)
\r
1664 sw::Shader::SourceParameter relativeRegister;
\r
1665 argument(relativeRegister, right);
\r
1667 dst.rel.index = relativeRegister.index;
\r
1668 dst.rel.type = relativeRegister.type;
\r
1669 dst.rel.scale = scale;
\r
1670 dst.rel.deterministic = !(vertexShader && left->getQualifier() == EvqUniform);
\r
1673 else if(dst.rel.index != registerIndex(&address)) // Move the previous index register to the address register
\r
1677 Constant oldScale((int)dst.rel.scale);
\r
1678 Instruction *mad = emit(sw::Shader::OPCODE_MAD, &address, &address, &oldScale, right);
\r
1679 mad->src[0].index = dst.rel.index;
\r
1680 mad->src[0].type = dst.rel.type;
\r
1684 Constant oldScale((int)dst.rel.scale);
\r
1685 Instruction *mul = emit(sw::Shader::OPCODE_MUL, &address, &address, &oldScale);
\r
1686 mul->src[0].index = dst.rel.index;
\r
1687 mul->src[0].type = dst.rel.type;
\r
1689 Constant newScale(scale);
\r
1690 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1693 dst.rel.type = sw::Shader::PARAMETER_TEMP;
\r
1694 dst.rel.index = registerIndex(&address);
\r
1695 dst.rel.scale = 1;
\r
1697 else // Just add the new index to the address register
\r
1701 emit(sw::Shader::OPCODE_ADD, &address, &address, right);
\r
1705 Constant newScale(scale);
\r
1706 emit(sw::Shader::OPCODE_MAD, &address, right, &newScale, &address);
\r
1710 else UNREACHABLE();
\r
1713 case EOpIndexDirectStruct:
\r
1715 const TTypeList *structure = left->getType().getStruct();
\r
1716 const TString &fieldName = right->getType().getFieldName();
\r
1719 for(TTypeList::const_iterator field = structure->begin(); field != structure->end(); field++)
\r
1721 if(field->type->getFieldName() == fieldName)
\r
1723 dst.type = registerType(left);
\r
1724 dst.index += offset;
\r
1725 dst.mask = writeMask(right);
\r
1730 offset += field->type->totalRegisterCount();
\r
1734 case EOpVectorSwizzle:
\r
1736 ASSERT(left->isRegister());
\r
1738 int leftMask = dst.mask;
\r
1741 int rightMask = 0;
\r
1743 TIntermSequence &sequence = right->getAsAggregate()->getSequence();
\r
1745 for(unsigned int i = 0; i < sequence.size(); i++)
\r
1747 int index = sequence[i]->getAsConstantUnion()->getIConst(0);
\r
1749 int element = swizzleElement(leftSwizzle, index);
\r
1750 rightMask = rightMask | (1 << element);
\r
1751 swizzle = swizzle | swizzleElement(leftSwizzle, i) << (element * 2);
\r
1754 dst.mask = leftMask & rightMask;
\r
1760 UNREACHABLE(); // Not an l-value operator
\r
1766 dst.type = registerType(symbol);
\r
1767 dst.index = registerIndex(symbol);
\r
1768 dst.mask = writeMask(symbol);
\r
1775 sw::Shader::ParameterType OutputASM::registerType(TIntermTyped *operand)
\r
1777 if(isSamplerRegister(operand))
\r
1779 return sw::Shader::PARAMETER_SAMPLER;
\r
1782 const TQualifier qualifier = operand->getQualifier();
\r
1783 if((EvqFragColor == qualifier) || (EvqFragData == qualifier))
\r
1785 if(((EvqFragData == qualifier) && (EvqFragColor == outputQualifier)) ||
\r
1786 ((EvqFragColor == qualifier) && (EvqFragData == outputQualifier)))
\r
1788 mContext.error(operand->getLine(), "static assignment to both gl_FragData and gl_FragColor", "");
\r
1790 outputQualifier = qualifier;
\r
1795 case EvqTemporary: return sw::Shader::PARAMETER_TEMP;
\r
1796 case EvqGlobal: return sw::Shader::PARAMETER_TEMP;
\r
1797 case EvqConstExpr: return sw::Shader::PARAMETER_FLOAT4LITERAL; // All converted to float
\r
1798 case EvqAttribute: return sw::Shader::PARAMETER_INPUT;
\r
1799 case EvqVaryingIn: return sw::Shader::PARAMETER_INPUT;
\r
1800 case EvqVaryingOut: return sw::Shader::PARAMETER_OUTPUT;
\r
1801 case EvqInvariantVaryingIn: return sw::Shader::PARAMETER_INPUT; // FIXME: Guarantee invariance at the backend
\r
1802 case EvqInvariantVaryingOut: return sw::Shader::PARAMETER_OUTPUT; // FIXME: Guarantee invariance at the backend
\r
1803 case EvqUniform: return sw::Shader::PARAMETER_CONST;
\r
1804 case EvqIn: return sw::Shader::PARAMETER_TEMP;
\r
1805 case EvqOut: return sw::Shader::PARAMETER_TEMP;
\r
1806 case EvqInOut: return sw::Shader::PARAMETER_TEMP;
\r
1807 case EvqConstReadOnly: return sw::Shader::PARAMETER_TEMP;
\r
1808 case EvqPosition: return sw::Shader::PARAMETER_OUTPUT;
\r
1809 case EvqPointSize: return sw::Shader::PARAMETER_OUTPUT;
\r
1810 case EvqFragCoord: return sw::Shader::PARAMETER_MISCTYPE;
\r
1811 case EvqFrontFacing: return sw::Shader::PARAMETER_MISCTYPE;
\r
1812 case EvqPointCoord: return sw::Shader::PARAMETER_INPUT;
\r
1813 case EvqFragColor: return sw::Shader::PARAMETER_COLOROUT;
\r
1814 case EvqFragData: return sw::Shader::PARAMETER_COLOROUT;
\r
1815 default: UNREACHABLE();
\r
1818 return sw::Shader::PARAMETER_VOID;
\r
1821 int OutputASM::registerIndex(TIntermTyped *operand)
\r
1823 if(isSamplerRegister(operand))
\r
1825 return samplerRegister(operand);
\r
1828 switch(operand->getQualifier())
\r
1830 case EvqTemporary: return temporaryRegister(operand);
\r
1831 case EvqGlobal: return temporaryRegister(operand);
\r
1832 case EvqConstExpr: UNREACHABLE();
\r
1833 case EvqAttribute: return attributeRegister(operand);
\r
1834 case EvqVaryingIn: return varyingRegister(operand);
\r
1835 case EvqVaryingOut: return varyingRegister(operand);
\r
1836 case EvqInvariantVaryingIn: return varyingRegister(operand);
\r
1837 case EvqInvariantVaryingOut: return varyingRegister(operand);
\r
1838 case EvqUniform: return uniformRegister(operand);
\r
1839 case EvqIn: return temporaryRegister(operand);
\r
1840 case EvqOut: return temporaryRegister(operand);
\r
1841 case EvqInOut: return temporaryRegister(operand);
\r
1842 case EvqConstReadOnly: return temporaryRegister(operand);
\r
1843 case EvqPosition: return varyingRegister(operand);
\r
1844 case EvqPointSize: return varyingRegister(operand);
\r
1845 case EvqFragCoord: pixelShader->vPosDeclared = true; return 0;
\r
1846 case EvqFrontFacing: pixelShader->vFaceDeclared = true; return 1;
\r
1847 case EvqPointCoord: return varyingRegister(operand);
\r
1848 case EvqFragColor: return 0;
\r
1849 case EvqFragData: return 0;
\r
1850 default: UNREACHABLE();
\r
1856 int OutputASM::writeMask(TIntermTyped *destination, int index)
\r
1858 if(destination->getQualifier() == EvqPointSize)
\r
1860 return 0x2; // Point size stored in the y component
\r
1863 return 0xF >> (4 - registerSize(destination->getType(), index));
\r
1866 int OutputASM::readSwizzle(TIntermTyped *argument, int size)
\r
1868 if(argument->getQualifier() == EvqPointSize)
\r
1870 return 0x55; // Point size stored in the y component
\r
1873 static const unsigned char swizzleSize[5] = {0x00, 0x00, 0x54, 0xA4, 0xE4}; // (void), xxxx, xyyy, xyzz, xyzw
\r
1875 return swizzleSize[size];
\r
1878 // Conservatively checks whether an expression is fast to compute and has no side effects
\r
1879 bool OutputASM::trivial(TIntermTyped *expression, int budget)
\r
1881 if(!expression->isRegister())
\r
1886 return cost(expression, budget) >= 0;
\r
1889 // Returns the remaining computing budget (if < 0 the expression is too expensive or has side effects)
\r
1890 int OutputASM::cost(TIntermNode *expression, int budget)
\r
1897 if(expression->getAsSymbolNode())
\r
1901 else if(expression->getAsConstantUnion())
\r
1905 else if(expression->getAsBinaryNode())
\r
1907 TIntermBinary *binary = expression->getAsBinaryNode();
\r
1909 switch(binary->getOp())
\r
1911 case EOpVectorSwizzle:
\r
1912 case EOpIndexDirect:
\r
1913 case EOpIndexDirectStruct:
\r
1914 return cost(binary->getLeft(), budget - 0);
\r
1918 return cost(binary->getLeft(), cost(binary->getRight(), budget - 1));
\r
1923 else if(expression->getAsUnaryNode())
\r
1925 TIntermUnary *unary = expression->getAsUnaryNode();
\r
1927 switch(unary->getOp())
\r
1931 return cost(unary->getOperand(), budget - 1);
\r
1936 else if(expression->getAsSelectionNode())
\r
1938 TIntermSelection *selection = expression->getAsSelectionNode();
\r
1940 if(selection->usesTernaryOperator())
\r
1942 TIntermTyped *condition = selection->getCondition();
\r
1943 TIntermNode *trueBlock = selection->getTrueBlock();
\r
1944 TIntermNode *falseBlock = selection->getFalseBlock();
\r
1945 TIntermConstantUnion *constantCondition = condition->getAsConstantUnion();
\r
1947 if(constantCondition)
\r
1949 bool trueCondition = constantCondition->getUnionArrayPointer()->getBConst();
\r
1953 return cost(trueBlock, budget - 0);
\r
1957 return cost(falseBlock, budget - 0);
\r
1962 return cost(trueBlock, cost(falseBlock, budget - 2));
\r
1970 const Function *OutputASM::findFunction(const TString &name)
\r
1972 for(unsigned int f = 0; f < functionArray.size(); f++)
\r
1974 if(functionArray[f].name == name)
\r
1976 return &functionArray[f];
\r
1983 int OutputASM::temporaryRegister(TIntermTyped *temporary)
\r
1985 return allocate(temporaries, temporary);
\r
1988 int OutputASM::varyingRegister(TIntermTyped *varying)
\r
1990 int var = lookup(varyings, varying);
\r
1994 var = allocate(varyings, varying);
\r
1995 int componentCount = varying->getNominalSize();
\r
1996 int registerCount = varying->totalRegisterCount();
\r
2000 if((var + registerCount) > sw::PixelShader::MAX_INPUT_VARYINGS)
\r
2002 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "fragment shader");
\r
2006 if(varying->getQualifier() == EvqPointCoord)
\r
2008 ASSERT(varying->isRegister());
\r
2009 if(componentCount >= 1) pixelShader->semantic[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2010 if(componentCount >= 2) pixelShader->semantic[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2011 if(componentCount >= 3) pixelShader->semantic[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2012 if(componentCount >= 4) pixelShader->semantic[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, var);
\r
2016 for(int i = 0; i < varying->totalRegisterCount(); i++)
\r
2018 if(componentCount >= 1) pixelShader->semantic[var + i][0] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2019 if(componentCount >= 2) pixelShader->semantic[var + i][1] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2020 if(componentCount >= 3) pixelShader->semantic[var + i][2] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2021 if(componentCount >= 4) pixelShader->semantic[var + i][3] = sw::Shader::Semantic(sw::Shader::USAGE_COLOR, var + i);
\r
2025 else if(vertexShader)
\r
2027 if((var + registerCount) > sw::VertexShader::MAX_OUTPUT_VARYINGS)
\r
2029 mContext.error(varying->getLine(), "Varyings packing failed: Too many varyings", "vertex shader");
\r
2033 if(varying->getQualifier() == EvqPosition)
\r
2035 ASSERT(varying->isRegister());
\r
2036 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2037 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2038 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2039 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_POSITION, 0);
\r
2040 vertexShader->positionRegister = var;
\r
2042 else if(varying->getQualifier() == EvqPointSize)
\r
2044 ASSERT(varying->isRegister());
\r
2045 vertexShader->output[var][0] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2046 vertexShader->output[var][1] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2047 vertexShader->output[var][2] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2048 vertexShader->output[var][3] = sw::Shader::Semantic(sw::Shader::USAGE_PSIZE, 0);
\r
2049 vertexShader->pointSizeRegister = var;
\r
2053 // Semantic indexes for user varyings will be assigned during program link to match the pixel shader
\r
2056 else UNREACHABLE();
\r
2058 declareVarying(varying, var);
\r
2064 void OutputASM::declareVarying(TIntermTyped *varying, int reg)
\r
2066 if(varying->getQualifier() != EvqPointCoord) // gl_PointCoord does not need linking
\r
2068 const TType &type = varying->getType();
\r
2069 const char *name = varying->getAsSymbolNode()->getSymbol().c_str();
\r
2070 VaryingList &activeVaryings = shaderObject->varyings;
\r
2072 // Check if this varying has been declared before without having a register assigned
\r
2073 for(VaryingList::iterator v = activeVaryings.begin(); v != activeVaryings.end(); v++)
\r
2075 if(v->name == name)
\r
2079 ASSERT(v->reg < 0 || v->reg == reg);
\r
2087 activeVaryings.push_back(glsl::Varying(glVariableType(type), name, varying->getArraySize(), reg, 0));
\r
2091 int OutputASM::uniformRegister(TIntermTyped *uniform)
\r
2093 const TType &type = uniform->getType();
\r
2094 ASSERT(!IsSampler(type.getBasicType()));
\r
2095 TIntermSymbol *symbol = uniform->getAsSymbolNode();
\r
2100 int index = lookup(uniforms, uniform);
\r
2104 index = allocate(uniforms, uniform);
\r
2105 const TString &name = symbol->getSymbol().c_str();
\r
2107 declareUniform(type, name, index);
\r
2116 int OutputASM::attributeRegister(TIntermTyped *attribute)
\r
2118 ASSERT(!attribute->isArray());
\r
2119 ASSERT(attribute->getBasicType() == EbtFloat);
\r
2121 int index = lookup(attributes, attribute);
\r
2125 TIntermSymbol *symbol = attribute->getAsSymbolNode();
\r
2130 index = allocate(attributes, attribute);
\r
2131 const TType &type = attribute->getType();
\r
2132 int registerCount = attribute->totalRegisterCount();
\r
2134 if(vertexShader && (index + registerCount) <= sw::VertexShader::MAX_INPUT_ATTRIBUTES)
\r
2136 for(int i = 0; i < registerCount; i++)
\r
2138 vertexShader->input[index + i] = sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i);
\r
2142 ActiveAttributes &activeAttributes = shaderObject->activeAttributes;
\r
2144 const char *name = symbol->getSymbol().c_str();
\r
2145 activeAttributes.push_back(Attribute(glVariableType(type), name, 0, index));
\r
2152 int OutputASM::samplerRegister(TIntermTyped *sampler)
\r
2154 ASSERT(IsSampler(sampler->getType().getBasicType()));
\r
2155 TIntermSymbol *symbol = sampler->getAsSymbolNode();
\r
2156 TIntermBinary *binary = sampler->getAsBinaryNode();
\r
2160 return samplerRegister(symbol);
\r
2164 ASSERT(binary->getOp() == EOpIndexDirect || binary->getOp() == EOpIndexIndirect || binary->getOp() == EOpIndexDirectStruct);
\r
2166 return samplerRegister(binary->getLeft()); // Index added later
\r
2168 else UNREACHABLE();
\r
2173 int OutputASM::samplerRegister(TIntermSymbol *sampler)
\r
2175 const TType &type = sampler->getType();
\r
2176 ASSERT(IsSampler(type.getBasicType()) || type.getStruct()); // Structures can contain samplers
\r
2178 int index = lookup(samplers, sampler);
\r
2182 index = allocate(samplers, sampler);
\r
2184 if(sampler->getQualifier() == EvqUniform)
\r
2186 const char *name = sampler->getSymbol().c_str();
\r
2187 declareUniform(type, name, index);
\r
2194 int OutputASM::lookup(VariableArray &list, TIntermTyped *variable)
\r
2196 for(unsigned int i = 0; i < list.size(); i++)
\r
2198 if(list[i] == variable)
\r
2200 return i; // Pointer match
\r
2204 TIntermSymbol *varSymbol = variable->getAsSymbolNode();
\r
2208 for(unsigned int i = 0; i < list.size(); i++)
\r
2212 TIntermSymbol *listSymbol = list[i]->getAsSymbolNode();
\r
2216 if(listSymbol->getId() == varSymbol->getId())
\r
2218 ASSERT(listSymbol->getSymbol() == varSymbol->getSymbol());
\r
2219 ASSERT(listSymbol->getType() == varSymbol->getType());
\r
2220 ASSERT(listSymbol->getQualifier() == varSymbol->getQualifier());
\r
2232 int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
\r
2234 int index = lookup(list, variable);
\r
2238 unsigned int registerCount = variable->totalRegisterCount();
\r
2240 for(unsigned int i = 0; i < list.size(); i++)
\r
2244 unsigned int j = 1;
\r
2245 for( ; j < registerCount && (i + j) < list.size(); j++)
\r
2247 if(list[i + j] != 0)
\r
2253 if(j == registerCount) // Found free slots
\r
2255 for(unsigned int j = 0; j < registerCount; j++)
\r
2257 list[i + j] = variable;
\r
2265 index = list.size();
\r
2267 for(unsigned int i = 0; i < registerCount; i++)
\r
2269 list.push_back(variable);
\r
2276 void OutputASM::free(VariableArray &list, TIntermTyped *variable)
\r
2278 int index = lookup(list, variable);
\r
2286 void OutputASM::declareUniform(const TType &type, const TString &name, int index)
\r
2288 const TTypeList *structure = type.getStruct();
\r
2289 ActiveUniforms &activeUniforms = shaderObject->activeUniforms;
\r
2293 activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(), index));
\r
2295 if(isSamplerRegister(type))
\r
2297 for(int i = 0; i < type.totalRegisterCount(); i++)
\r
2299 shader->declareSampler(index + i);
\r
2305 if(type.isArray())
\r
2307 int elementIndex = index;
\r
2309 for(int i = 0; i < type.getArraySize(); i++)
\r
2311 for(size_t j = 0; j < structure->size(); j++)
\r
2313 const TType &fieldType = *(*structure)[j].type;
\r
2314 const TString &fieldName = fieldType.getFieldName();
\r
2316 const TString uniformName = name + "[" + str(i) + "]." + fieldName;
\r
2317 declareUniform(fieldType, uniformName, elementIndex);
\r
2318 elementIndex += fieldType.totalRegisterCount();
\r
2324 int fieldIndex = index;
\r
2326 for(size_t i = 0; i < structure->size(); i++)
\r
2328 const TType &fieldType = *(*structure)[i].type;
\r
2329 const TString &fieldName = fieldType.getFieldName();
\r
2331 const TString uniformName = name + "." + fieldName;
\r
2332 declareUniform(fieldType, uniformName, fieldIndex);
\r
2333 fieldIndex += fieldType.totalRegisterCount();
\r
2339 GLenum OutputASM::glVariableType(const TType &type)
\r
2341 if(type.getBasicType() == EbtFloat)
\r
2343 if(type.isScalar())
\r
2347 else if(type.isVector())
\r
2349 switch(type.getNominalSize())
\r
2351 case 2: return GL_FLOAT_VEC2;
\r
2352 case 3: return GL_FLOAT_VEC3;
\r
2353 case 4: return GL_FLOAT_VEC4;
\r
2354 default: UNREACHABLE();
\r
2357 else if(type.isMatrix())
\r
2359 switch(type.getNominalSize())
\r
2362 switch(type.getSecondarySize())
\r
2364 case 2: return GL_FLOAT_MAT2;
\r
2365 case 3: return GL_FLOAT_MAT2x3;
\r
2366 case 4: return GL_FLOAT_MAT2x4;
\r
2367 default: UNREACHABLE();
\r
2370 switch(type.getSecondarySize())
\r
2372 case 2: return GL_FLOAT_MAT3x2;
\r
2373 case 3: return GL_FLOAT_MAT3;
\r
2374 case 4: return GL_FLOAT_MAT4x2;
\r
2375 default: UNREACHABLE();
\r
2378 switch(type.getSecondarySize())
\r
2380 case 2: return GL_FLOAT_MAT4x2;
\r
2381 case 3: return GL_FLOAT_MAT4x3;
\r
2382 case 4: return GL_FLOAT_MAT4;
\r
2383 default: UNREACHABLE();
\r
2385 default: UNREACHABLE();
\r
2388 else UNREACHABLE();
\r
2390 else if(type.getBasicType() == EbtInt)
\r
2392 if(type.isScalar())
\r
2396 else if(type.isVector())
\r
2398 switch(type.getNominalSize())
\r
2400 case 2: return GL_INT_VEC2;
\r
2401 case 3: return GL_INT_VEC3;
\r
2402 case 4: return GL_INT_VEC4;
\r
2403 default: UNREACHABLE();
\r
2406 else UNREACHABLE();
\r
2408 else if(type.getBasicType() == EbtBool)
\r
2410 if(type.isScalar())
\r
2414 else if(type.isVector())
\r
2416 switch(type.getNominalSize())
\r
2418 case 2: return GL_BOOL_VEC2;
\r
2419 case 3: return GL_BOOL_VEC3;
\r
2420 case 4: return GL_BOOL_VEC4;
\r
2421 default: UNREACHABLE();
\r
2424 else UNREACHABLE();
\r
2426 else if(type.getBasicType() == EbtSampler2D)
\r
2428 return GL_SAMPLER_2D;
\r
2430 else if(type.getBasicType() == EbtSamplerCube)
\r
2432 return GL_SAMPLER_CUBE;
\r
2434 else if(type.getBasicType() == EbtSamplerExternalOES)
\r
2436 return GL_SAMPLER_EXTERNAL_OES;
\r
2438 else if(type.getBasicType() == EbtSampler3D)
\r
2440 return GL_SAMPLER_3D_OES;
\r
2442 else UNREACHABLE();
\r
2447 GLenum OutputASM::glVariablePrecision(const TType &type)
\r
2449 if(type.getBasicType() == EbtFloat)
\r
2451 switch(type.getPrecision())
\r
2453 case EbpHigh: return GL_HIGH_FLOAT;
\r
2454 case EbpMedium: return GL_MEDIUM_FLOAT;
\r
2455 case EbpLow: return GL_LOW_FLOAT;
\r
2456 case EbpUndefined:
\r
2457 // Should be defined as the default precision by the parser
\r
2458 default: UNREACHABLE();
\r
2461 else if(type.getBasicType() == EbtInt)
\r
2463 switch (type.getPrecision())
\r
2465 case EbpHigh: return GL_HIGH_INT;
\r
2466 case EbpMedium: return GL_MEDIUM_INT;
\r
2467 case EbpLow: return GL_LOW_INT;
\r
2468 case EbpUndefined:
\r
2469 // Should be defined as the default precision by the parser
\r
2470 default: UNREACHABLE();
\r
2474 // Other types (boolean, sampler) don't have a precision
\r
2478 int OutputASM::dim(TIntermNode *v)
\r
2480 TIntermTyped *vector = v->getAsTyped();
\r
2481 ASSERT(vector && vector->isRegister());
\r
2482 return vector->getNominalSize();
\r
2485 int OutputASM::dim2(TIntermNode *m)
\r
2487 TIntermTyped *matrix = m->getAsTyped();
\r
2488 ASSERT(matrix && matrix->isMatrix() && !matrix->isArray());
\r
2489 return matrix->getNominalSize();
\r
2492 // Returns ~0 if no loop count could be determined
\r
2493 unsigned int OutputASM::loopCount(TIntermLoop *node)
\r
2495 // Parse loops of the form:
\r
2496 // for(int index = initial; index [comparator] limit; index += increment)
\r
2497 TIntermSymbol *index = 0;
\r
2498 TOperator comparator = EOpNull;
\r
2501 int increment = 0;
\r
2503 // Parse index name and intial value
\r
2504 if(node->getInit())
\r
2506 TIntermAggregate *init = node->getInit()->getAsAggregate();
\r
2510 TIntermSequence &sequence = init->getSequence();
\r
2511 TIntermTyped *variable = sequence[0]->getAsTyped();
\r
2513 if(variable && variable->getQualifier() == EvqTemporary)
\r
2515 TIntermBinary *assign = variable->getAsBinaryNode();
\r
2517 if(assign->getOp() == EOpInitialize)
\r
2519 TIntermSymbol *symbol = assign->getLeft()->getAsSymbolNode();
\r
2520 TIntermConstantUnion *constant = assign->getRight()->getAsConstantUnion();
\r
2522 if(symbol && constant)
\r
2524 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2527 initial = constant->getUnionArrayPointer()[0].getIConst();
\r
2535 // Parse comparator and limit value
\r
2536 if(index && node->getCondition())
\r
2538 TIntermBinary *test = node->getCondition()->getAsBinaryNode();
\r
2540 if(test && test->getLeft()->getAsSymbolNode()->getId() == index->getId())
\r
2542 TIntermConstantUnion *constant = test->getRight()->getAsConstantUnion();
\r
2546 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2548 comparator = test->getOp();
\r
2549 limit = constant->getUnionArrayPointer()[0].getIConst();
\r
2555 // Parse increment
\r
2556 if(index && comparator != EOpNull && node->getExpression())
\r
2558 TIntermBinary *binaryTerminal = node->getExpression()->getAsBinaryNode();
\r
2559 TIntermUnary *unaryTerminal = node->getExpression()->getAsUnaryNode();
\r
2561 if(binaryTerminal)
\r
2563 TOperator op = binaryTerminal->getOp();
\r
2564 TIntermConstantUnion *constant = binaryTerminal->getRight()->getAsConstantUnion();
\r
2568 if(constant->getBasicType() == EbtInt && constant->getNominalSize() == 1)
\r
2570 int value = constant->getUnionArrayPointer()[0].getIConst();
\r
2574 case EOpAddAssign: increment = value; break;
\r
2575 case EOpSubAssign: increment = -value; break;
\r
2576 default: UNIMPLEMENTED();
\r
2581 else if(unaryTerminal)
\r
2583 TOperator op = unaryTerminal->getOp();
\r
2587 case EOpPostIncrement: increment = 1; break;
\r
2588 case EOpPostDecrement: increment = -1; break;
\r
2589 case EOpPreIncrement: increment = 1; break;
\r
2590 case EOpPreDecrement: increment = -1; break;
\r
2591 default: UNIMPLEMENTED();
\r
2596 if(index && comparator != EOpNull && increment != 0)
\r
2598 if(comparator == EOpLessThanEqual)
\r
2600 comparator = EOpLessThan;
\r
2604 if(comparator == EOpLessThan)
\r
2606 int iterations = (limit - initial) / increment;
\r
2608 if(iterations <= 0)
\r
2613 return iterations;
\r
2615 else UNIMPLEMENTED(); // Falls through
\r
2621 bool DetectLoopDiscontinuity::traverse(TIntermNode *node)
\r
2624 loopDiscontinuity = false;
\r
2626 node->traverse(this);
\r
2628 return loopDiscontinuity;
\r
2631 bool DetectLoopDiscontinuity::visitLoop(Visit visit, TIntermLoop *loop)
\r
2633 if(visit == PreVisit)
\r
2637 else if(visit == PostVisit)
\r
2645 bool DetectLoopDiscontinuity::visitBranch(Visit visit, TIntermBranch *node)
\r
2647 if(loopDiscontinuity)
\r
2657 switch(node->getFlowOp())
\r
2664 loopDiscontinuity = true;
\r
2666 default: UNREACHABLE();
\r
2669 return !loopDiscontinuity;
\r
2672 bool DetectLoopDiscontinuity::visitAggregate(Visit visit, TIntermAggregate *node)
\r
2674 return !loopDiscontinuity;
\r